| Complex adaptive system is usually composed of many parallel and adaptive agents and provided with hierarchical structure, intelligence and self-adaptability. Emergence is one typical feature of such systems. So it is hard to parse and validate complex adaptive sysem by formalization. Simulation is a feasible approach to this goal, which can formally describe and simulate these systems using distributed and collaborative models. It is important for domain experts to study behaviors and characteristics of a system and to forecast its changes and trends. Similarly, it is significant for a skilled person to engage in operational action, training and rehearsing within the system.At present, agent/swarm is regarded as basic unit in modeling and simulation based on the theory of complex adaptive system, which makes it convenient to model small scale systems and keep high simulation speed. But with the increasing of system scale, interactions among numerous of agents pose higher requirements for the simulation. So the distributed and interactive simulation must be further perfected. Swarm is based on event queue, which can not adequately simulate the randomness of event among simulation agents. Thus, more efficient method is needed to implement the collaboration among simulation objects. Traditional mathematics models can not properly describe changes of complex situation and evolutionary mechanism. Thereby, the dynamics of system should be further depicted. The individual autonomy of agent is mainly interested in swarm. However, centralized control and horizontal ties for different groups are not described in depth. Consequently, the hierarchy and modularity of system ought to be intensively depicted.In order to solve the above-mentioned problems, a technical framework of combining modeling and simulation was proposed by use of the fundamental theory of complex adaptive system. The modeling method based on Agent was introduced into DEVS formalism, which has the advantages of hierarchy and modularity. As a result, this formalism was extended to Agent-DEVS formalism. The distributed and interactive co-simulation was carried out based on HLA. And the parameter optimization of Agent-DEVS model was accomplished based on SVM. Ultimately, a prototype system of simulation training for material supply in emergent disaster was constructed. Generally, main contributions of this thesis are shown as follows: â‘ Agent-DEVS, an extended DEVS formalism with description capabilities of intelligence and cooperation, was proposed. It was extended based on Parallel-DEVS. State element was expanded to Agent personality element. Agent model element was added to represent intelligent individual behavior. And the input and output of model ports were extended to Agent message to represent social cooperation. Formal specifications of Agent-DEVS atomic model and coupled model were analyzed. The closure of Agent-DEVS models under coupling was proved. And an implementation algorithm of Agent-DEVS models was given. The results of simulation test show that: 1) Agent-DEVS formalism can describe intelligent behavior directly. Main merit of Agent-DEVS is modeling performance. It translates all kinds of parameters into variables stored in knowledge and constructs relevant processing function set, which can describe more complex intelligent behavior. 2) Agent-DEVS formalism can properly describe cooperative behavior. It may dynamically modify the information in knowledge through mutual cooperation among models, which improves the abilities of models for dealing with transactions independently and enhances the autonomy of models. 3) The implementation algorithm of Agent-DEVS models is by no means inferior in the time complexity and modeling efficiency. If considered in computational time, this algorithm only increases the steps for Agent operations in inmost atomic models, which has less impact on the whole complexity and runtime. In terms of modeling efficiency, it is equivalent to that of DEVS/CD++.â‘¡Collaborative modeling and simulation method for Agent-DEVS based on HLA was studied. Formal description of Agent-DEVS federate models based on HLA was proposed. All ports of Agent-DEVS models were translated into data objects of HLA. Structure of Agent-DEVS federate models was showed and communication mechanism was analyzed. The mapping between knowledge update in Agent-DEVS and attribute update in HLA, and that between models coupling in Agent-DEVS and instances interaction in HLA, were established respectively. Simulation flow of Agent-DEVS federate models in HLA was designed. The results of simulation test show that: 1) HLA enhances the reusability of Agent-DEVS models. Distributed interaction between Agent-DEVS models and that between Agent-DEVS model and non Agent-DEVS model may be implemented at the same time. 2) Knowledge update mechanism enriches the interoperability of Agent-DEVS models. The update of knowledge repository is separated from interaction of models coupling, which may avoid bulk data transfering in coupling interactions and improve update efficiency of knowledge repository. 3) Independent ability for transaction processing of Agent-DEVS models is improved by real-time knowledge. And the autonomy of Agent-DEVS models is enhanced significantly.â‘¢Optimization model for parameters in Agent-DEVS models based on SVM was constructed. Parameter optimization flow of Agent-DEVS models based on SVM was put forward. And the optimization model for main parameter in simulation models of material supply in emergent disaster was established. Some key techniques, such as collection and pretreatment of data, selection of kernel function and optimization selection of SVM parameters, were mainly analyzed. And the comparison with BP neural networks was examined. The results of simulation test show that: 1) SVM enhances the dynamics of Agent-DEVS models. Parameters in the models can be appropriately forecasted by a few samples. So the objective of parameter optimization is reached. 2) Self-learning ability improves the intelligence of Agent-DEVS models significantly. Method of machine learning may handle more complex intelligent behavior. With further application of models and accumulation of samples, the forecast ability will be advanced automatically. It is provided with strong autonomy. 3) Optimized parameters make Agent-DEVS models more elaborate descriptive ability. These parameters are regarded as running basis of models, which may raise the computational precision and increase the level of detail in descriptive object.â‘£The technology of combining modeling and simulation was applied to simulation training in emergent disaster, and the prototype system of simulation training for material supply in emergent disaster was constructed. Construction and performance objectives were brought forward. Operation details and processes of material supply in emergent disaster were analyzed. Then function requirements of the system were proposed. A three-layered system structure was expatiated on. Designing flow of simulation models based on technology of combining modeling and simulation was given. And distributed system design based on HLA was completed. Evaluation indicator system and fuzzy comprehensive evaluation model for simulation training system were presented respectively. And the subsystem of simulation training for camping material supply in emergent disaster was estimated and analyzed. The results of evaluation show that the performance and quality grade of this subsystem is good.
|
PDF Full Text Request